The peak-end rule says people judge a past experience mostly by how it felt at its most intense moment and at its end, not by the total or average of every moment.
Daniel Kahneman and colleagues showed it in pain experiments in the 1990s, and a 2022 meta-analysis of 174 effect sizes found the effect is large and robust.
For stores and apps, it means the checkout finish, the order confirmation and the delivery moment carry more weight in memory than their share of the journey suggests.
This page covers what the research found, where it falls short, and what real A/B tests in our test database show when teams design for peaks and endings.
One honest caveat up front: no test we have measures remembered experience directly.
They measure conversion, revenue and acceptance rates, so they tell you about behavior in the moment, not about memory.
What the peak-end rule is
When people look back on an episode, they don't add up how it felt minute by minute.
They form a summary from a few snapshots, and two snapshots dominate: the peak (the best or worst moment) and the end.
A second finding usually comes with it: duration neglect.
How long the experience lasted barely changes how people rate it afterward.
Barbara Fredrickson and Daniel Kahneman named this in a 1993 study of people rating film clips in the Journal of Personality and Social Psychology.
The rule is one of many cognitive biases about memory and judgment.
It sits close to the serial position effect, which says the first and last items in a list are remembered best.
The difference is that peak-end is about how people feel about an experience, not what they recall from it.
The research behind it
The cold-water experiment (1993)
In "When More Pain Is Preferred to Less: Adding a Better End", Kahneman, Fredrickson, Charles Schreiber and Donald Redelmeier had 32 male university students put one hand in 14°C water for 60 seconds.
In a second trial, the other hand stayed in 14°C water for 60 seconds, then 30 more seconds while the water was slowly warmed to 15°C.
The long trial contained all the pain of the short one plus more.
Yet when asked which trial to repeat, 22 of 32 subjects (69%) chose the long one.
In a replication without real-time pain ratings, 24 of 37 participants chose the long trial.
The colonoscopy trial (2003)
The lab result held up in a hospital.
In a randomized trial published in Pain, Redelmeier, Joel Katz and Kahneman studied 682 colonoscopy patients.
Half had a short, milder interval added to the end of their procedure.
That group rated the final moments as less painful (1.7 vs. 2.5 on a 10-point scale), the whole procedure as less unpleasant (4.4 vs. 4.9 on a 10 cm scale), and were somewhat more likely to come back for a repeat colonoscopy over a median 5.3 years of follow-up (odds ratio 1.41).
The longer procedure produced a better memory and a small change in future behavior.
Pleasant experiences: the DVD study (2008)
Most early work was on pain. Amy Do, Alexander Rupert and George Wolford tested pleasure in a 2008 paper in Psychonomic Bulletin & Review.
Donors who won a raffle picked a free DVD from a list of highly rated films; some then got a second pick from a list of good but less acclaimed films.
Across 100 participants, people who got one great DVD were more pleased with the whole experience than people who got the great DVD followed by a lesser one.
A good thing added at the end made the overall memory worse, because it lowered the ending.
People who got the lesser DVD first and the great one second did not rate the experience lower than the one-great-DVD group.
What the research does not show
The rule is well supported, but it is often stretched. Three limits matter for anyone planning tests.
- The middle still counts. The 2022 meta-analysis by Balca Alaybek and colleagues in Organizational Behavior and Human Decision Processes found a large peak-end effect (r = 0.581, 95% CI 0.487 to 0.661). It also found that effect was comparable to the simple average of all moments. Peak and end are a good shortcut, not a license to neglect everything in between.
- Small, specific samples. The famous cold-water study used 32 male students. The colonoscopy trial is far stronger evidence, and the effect there on future behavior was modest.
- Memory is not conversion. These studies measured ratings and later choices. A better-remembered checkout should, in theory, help repeat purchase and reviews, but that link is rarely tested in ecommerce. Treat it as a hypothesis, not a guaranteed lift.
The meta-analysis is the best reason to take the rule seriously: across many settings, peaks and ends predicted overall ratings well, and duration predicted almost nothing.
Where the peaks and ends are in a customer journey
The "experience" a customer remembers depends on how they frame the episode.
A shopping session, an order, and a whole customer relationship each have their own peak and end.
| Episode | Likely peak | Likely end | What you can measure in a test |
|---|---|---|---|
| Shopping session | Finding the right product, picking a bundle | Payment and order confirmation | Conversion, revenue per visitor |
| Order | Delivery and unboxing, first use | Delivery, or the first problem with it | Repeat purchase rate, review rate, support tickets |
| App or SaaS trial | First real value ("aha" moment) | Trial end, paywall, or cancellation | Activation, trial conversion, win-back rate |
| Subscription | Best month of use | Cancellation flow | Reactivation rate, NPS of churned users |
Endings are often the weakest part of a journey because teams optimize the steps before the sale.
A clunky thank-you page, a confirmation email full of upsells, or a hostile cancellation flow are all endings.
What our A/B tests show about peaks
The closest test in our database to a designed "peak" is a confetti test credited to Dave Diederen.
When a shopper selected the largest bundle on a product page, the store triggered a confetti effect and a message saying they had unlocked the biggest savings.
It won with +5.47% conversion rate and +6.32% revenue per visitor.
That is a positive moment placed at the point of a meaningful decision.
It shows that a small moment of reward can change behavior right then.
It doesn't show anything about memory, since the metric was the purchase in the same session.
On the SaaS side, AURUM (Astrea) changed its trial so new users saw real legal clippings on their first login instead of waiting up to 48 hours.
Engagement with a clipping rose by 16 percentage points (+125% relative).
Moving the first moment of real value earlier is a way of creating a peak while the user is still deciding whether the product is worth it.
For more on this, see the goal-gradient and Zeigarnik effects.
What our A/B tests show about endings
Post-purchase offers
The page right after payment is the literal end of the shopping episode, and most tests there are upsells.
Dave Diederen reports a 1-click complementary post-purchase upsell with a 16.64% offer acceptance rate and $16.50 per accepted offer.
Visually reports that The Perfect Jean moved from offering a discounted hoodie to the item just bought in a new color, and offer conversion went from 1.8% to 18.6%.
That figure compares many iterations across tools, not one clean split test.
Vasta reports a post-purchase upsell that added $5,282.64 in its first week.
The offer itself isn't described and the monthly figure looks extrapolated.
These results fit the DVD study in an interesting way.
Both successful offers were closely matched to what the customer had just bought, so for a buyer who wants it, the ending gets better.
A poorly matched offer is the "lesser DVD" case: it adds something to the end that lowers it.
None of these tests measured satisfaction, returns or repeat purchase, so the effect on memory is unknown.
Our guide to upselling covers offer design in more depth.
Asking for a rating at the right moment
A rating prompt asks people to summarize their experience, which is exactly the judgment the peak-end rule describes.
OnSwitch tested an app rating pop-up that used heart and broken-heart emoji instead of text options, and got +14.65% rate taps and -2.23% skips.
The research suggests a second lever this test didn't touch: timing.
A prompt shown right after a success (a finished workout, a delivered order) catches people at a good end.
A prompt shown after an error catches them at a bad one.
If you run this test, compare timing with a post-purchase survey or review request sent after delivery rather than right after payment.
Cancellation and win-back
A cancellation flow is how a subscription ends, and it shapes whether someone comes back.
Unscripted showed churned users a personal paywall with their name, an extra 7-day free trial and a renewal timeline.
Trial starts rose by over 240%, trial conversion by over 30%, and proceeds per user nearly doubled.
The flip side is that forced, hard-to-exit cancellation flows end the relationship on its worst moment.
That is a dark pattern, and the peak-end rule predicts it makes win-back harder.
See our churn reduction guide for tested alternatives.
@DaveDiederen
Visually@DaveDiederenHow to test the peak-end rule properly
Most teams who cite the rule never test the part that makes it interesting: memory and what people do next. A few practical rules help.
- Pick a metric that happens later. Repeat purchase within 60 or 90 days, review rate, support contact rate or reactivation rate. Same-session conversion only tells you about the moment.
- Change the ending, not the whole flow. The colonoscopy trial changed only the last minutes. If you change several steps at once, you can't say whether the end caused the result.
- Watch the cost of a weaker end. A post-purchase offer can raise revenue per order while lowering satisfaction. Track refunds and returns on orders that saw the offer.
- Expect small effects. In the colonoscopy trial, the effect on return visits was real but modest. Plan sample sizes and test length for that.
- Don't ignore the middle. The meta-analysis found the average of all moments predicted ratings about as well as peak and end. Fixing a painful checkout step still matters.
For related ideas on how people judge choices and outcomes, browse the rest of our psychology hub.